Mailman 3 December 1999 - Python-list

TopSort in Python?
by Tim Peters Jan. 21, 2008

Jan. 21, 2008

From: "Tim Peters" <tim_one(a)email.msn.com> [Dinu C. Gherman] > Does anybody have a simple-minded-but-working full-Python > implementation of topsort, the topological sorting algorithm? > Or maybe *any* topsort? I remember only one such algorithm... > python.org/search doesn't reveal any... Searching for "topological" instead turns up two hits there, a few more on DejaNews. Apparently "the std" algorithm I posted years ago predates DejaNews, though. So here's a fancier version ... check-out-aaron's-kjbuckets-too-ly y'rs - tim # topsort takes a list of pairs, where each pair (x, y) is taken to # mean that x <= y wrt some abstract partial ordering. The return # value is a list, representing a total ordering that respects all # the input constraints. # E.g., # topsort( [(1,2), (3,3)] ) # may return any of (but nothing other than) # [3, 1, 2] # [1, 3, 2] # [1, 2, 3] # because those are the permutations of the input elements that # respect the "1 precedes 2" and "3 precedes 3" input constraints. # Note that a constraint of the form (x, x) is really just a trick # to make sure x appears *somewhere* in the output list. # # If there's a cycle in the constraints, say # topsort( [(1,2), (2,1)] ) # then CycleError is raised, and the exception object supports # many methods to help analyze and break the cycles. This requires # a good deal more code than topsort itself! from exceptions import Exception class CycleError(Exception): def __init__(self, sofar, numpreds, succs): Exception.__init__(self, "cycle in constraints", sofar, numpreds, succs) self.preds = None # return as much of the total ordering as topsort was able to # find before it hit a cycle def get_partial(self): return self[1] # return remaining elt -> count of predecessors map def get_pred_counts(self): return self[2] # return remaining elt -> list of successors map def get_succs(self): return self[3] # return remaining elements (== those that don't appear in # get_partial()) def get_elements(self): return self.get_pred_counts().keys() # Return a list of pairs representing the full state of what's # remaining (if you pass this list back to topsort, it will raise # CycleError again, and if you invoke get_pairlist on *that* # exception object, the result will be isomorphic to *this* # invocation of get_pairlist). # The idea is that you can use pick_a_cycle to find a cycle, # through some means or another pick an (x,y) pair in the cycle # you no longer want to respect, then remove that pair from the # output of get_pairlist and try topsort again. def get_pairlist(self): succs = self.get_succs() answer = [] for x in self.get_elements(): if succs.has_key(x): for y in succs[x]: answer.append( (x, y) ) else: # make sure x appears in topsort's output! answer.append( (x, x) ) return answer # return remaining elt -> list of predecessors map def get_preds(self): if self.preds is not None: return self.preds self.preds = preds = {} remaining_elts = self.get_elements() for x in remaining_elts: preds[x] = [] succs = self.get_succs() for x in remaining_elts: if succs.has_key(x): for y in succs[x]: preds[y].append(x) if __debug__: for x in remaining_elts: assert len(preds[x]) > 0 return preds # return a cycle [x, ..., x] at random def pick_a_cycle(self): remaining_elts = self.get_elements() # We know that everything in remaining_elts has a predecessor, # but don't know that everything in it has a successor. So # crawling forward over succs may hit a dead end. Instead we # crawl backward over the preds until we hit a duplicate, then # reverse the path. preds = self.get_preds() from random import choice x = choice(remaining_elts) answer = [] index = {} in_answer = index.has_key while not in_answer(x): index[x] = len(answer) # index of x in answer answer.append(x) x = choice(preds[x]) answer.append(x) answer = answer[index[x]:] answer.reverse() return answer def topsort(pairlist): numpreds = {} # elt -> # of predecessors successors = {} # elt -> list of successors for first, second in pairlist: # make sure every elt is a key in numpreds if not numpreds.has_key(first): numpreds[first] = 0 if not numpreds.has_key(second): numpreds[second] = 0 # if they're the same, there's no real dependence if first == second: continue # since first < second, second gains a pred ... numpreds[second] = numpreds[second] + 1 # ... and first gains a succ if successors.has_key(first): successors[first].append(second) else: successors[first] = [second] # suck up everything without a predecessor answer = filter(lambda x, numpreds=numpreds: numpreds[x] == 0, numpreds.keys()) # for everything in answer, knock down the pred count on # its successors; note that answer grows *in* the loop for x in answer: assert numpreds[x] == 0 del numpreds[x] if successors.has_key(x): for y in successors[x]: numpreds[y] = numpreds[y] - 1 if numpreds[y] == 0: answer.append(y) # following "del" isn't needed; just makes # CycleError details easier to grasp del successors[x] if numpreds: # everything in numpreds has at least one predecessor -> # there's a cycle if __debug__: for x in numpreds.keys(): assert numpreds[x] > 0 raise CycleError(answer, numpreds, successors) return answer

8 9

Dealing with faults matters (was: Is Python dying?)
by claird＠Starbase.NeoSoft.COM Aug. 28, 2003

Aug. 28, 2003

In article <3706f449.12085377(a)news.cybercity.dk>, Morten Christensen <mmc(a)vip.cybercity.dk> wrote: >On 31 Mar 1999 06:01:03 -0600, claird(a)Starbase.NeoSoft.COM (Cameron >Laird) wrote: >>*I* certainly discuss it. Anyone who understands >>that exception-handling is important I invite to >>realize that exception-handling on the sociologi- >>cal-organizational dimension has an even bigger >>multiplier. >> >>This is a plank of my "Visual C considered profes- >>sional malpractice" campaign. > >??? > >...I understand exception handling and regard it as very important, >although the (Visual) C++ implementation is really awful (as everyone >who have experienced "unexplainable" Visual C++ exceptions of unknown >origin and type knows ).... I also know a lot about how exception >handling is actually implemented in modern OO languages such as C++ >(this was the subject I wrote my master thesis on).... > >HOWEVER, I have NO clue about what you mean by exception-handling as a >multiplier on the sociological-organization dimension :-) I would like >to find out. Can you explain? . . . Happily. Excuse my giddiness, please; I see that that last message *did* wander too far into the land of obscurity. Here's what I was saying: exception-handling is important. Programs that intelligently use exception-handling mechanisms are better programs. Exception-handling is even more important to people. People and organizations that deal with exceptions intelligently, and don't pre- tend that everything should be without fault, are better people and organizations. In fact, a correct computer program often has over half its lines devoted to accomo- dating error. A wise person will have good ways to respond to criticism/problems/diffi- culties/...; dealing with success is relatively easier. -- Cameron Laird http://starbase.neosoft.com/~claird/home.html claird(a)NeoSoft.com +1 281 996 8546 FAX

3 2

Bug in v1.5.2 Websucker GUI?
by dhfx＠realtime.net June 13, 2003

June 13, 2003

I'm running v1.5.2 on a Win95 box. I tried wsgui.py and got the following error message: ... File "C:\ ... \WSGUI.PY", line 165, in go self.sucker.rootdir = os.path.dirname( [ websucker.Sucker.savefilename(self.url)) ] TypeError: unbound method must be called with class instance 1st arg (the part in [ ] didn't appear in the message - I added it from the code) The object-instance self.sucker is an instantiation of class SuckerThread, which is derived from class Sucker of the websucker code module. The SuckerThread class includes a method savefilename() which overwrites the method in websucker.Sucker of the same name. Whoever coded wsgui evidently wanted to use the method of the parent class, rather than the method of the derived class. What seems to be giving trouble is that the arg of dirname() is a class method, rather than a method of an instance of the class. Here's what I did: I defined a new method savefilename1() in SuckerThread which just calls the savefilename() of the parent class: def savefilename1(self, url): return websucker.Sucker.savefilename(self, url) Then I changed the code at "line 165" to the following: self.sucker.rootdir = os.path.dirname( self.sucker.savefilename1(url)) and now it works. Note that the arg "self" of savefilename1() in the line above is not needed, it's implicit. In fact, when I had it in, I got an error message that I had too many args! David Friedman

5 8

Re: automatic "lotjes trekken"
by François Pinard Aug. 26, 2002

Aug. 26, 2002

"Tim Peters" <tim_one(a)email.msn.com> writes: > [...] with a belly full of far too much turkey. Strange. I was imagining you rather slim. Are you? :-) > where p[:] ranges over all permutations of range(N) [...] By the way, I do not remember any standard library helper to enumerate all permutations (or combinations, or arrangements) of a given list (or tuple). Ideally one at a time, of course. Would this be useful? -- François Pinard http://www.iro.umontreal.ca/~pinard

4 5

pytags ?????
by Keith Dart Jan. 28, 2000

Jan. 28, 2000

Is anyone aware of a program that generates ctags-like tags file, except for Python code? I guess it would be named something like pytags (ptags is already taken by perl-tags and prolog-tags). I can't seem to find one. /___\\|//__//// \ (+ +) \\\\ -- --------------------oOOo~(_)~oOOo---------------------------------------- Keith Dart <mailto:kdart@techie.com> <http://www.employees.org/~kdart/> ============================================================================

4 3

Speed issues
by Gerrit Holl Jan. 16, 2000

Jan. 16, 2000

Hello, is an assigment slower than appending to a list? If I've got this function: def f(s): o=[] for c in s: o.append(chr(ord(c)/2)) return string.join(o, '') or: def f(s): o='' for c in s: o = o + chr(ord(c)/2) return o What is faster? What do you prefer? regards, Gerrit. -- "However, complexity is not always the enemy." -- Larry Wall (Open Sources, 1999 O'Reilly and Associates) 2:25pm up 4:51, 12 users, load average: 0.23, 0.11, 0.03

5 4

Python not a Very High-Level Language?
by Ajith Prasad Jan. 16, 2000

Jan. 16, 2000

http://www.oreilly.com/news/vhll_1299.html is an article by Greg Wilson casting doubts on the effectiveness/value of Python and other very high level scripting languages. Wilson comments that: "Over the past few years, I have done several medium-sized projects using both Perl and Python. At first, I was very excited by what these Very-High Level Languages (VHLLs) let me do, and how quickly. The more I played with them, however, the less satisfied I was. In particular, I no longer believe that they deserve the 'V' in their name. This article explores why, and suggests some ways in which they could evolve." Worth responding to as it includes detailed criticisms of Python in particular.

30 45

Python native compiler on NT
by tpchang＠excite.com Jan. 14, 2000

Jan. 14, 2000

I wonder if there is a Python native compiler for NT. This will help me to sell Python to people in my company who are more familiar with .exe files than the .py files :-) Thanks in advance, Ben Sent via Deja.com http://www.deja.com/ Before you buy.

5 4

FINANCIAL MODELLING PACKAGE IN PYTHON?
by Ajith Prasad Jan. 9, 2000

Jan. 9, 2000

Andy Robinson's website at http://www.robanal.demon.co.uk mentions his work on DoubleTalk which uses Python to handle accounting transactions. I saw also references to work by Kevin McDermott on use of Python in finance. I wonder whether anyone is working on a full-fledged financial modelling package in Python? I have in mind a package similar to IFPS (Interactive Financial Planning System) described in "Visual IFPS/Plus for Business" by Paul Gray (details at http://www.amazon.com), but extensible in Python with OLAP-type functionality. I am sure that there would be a good market for such a product among finance professionals who need much more than a spreadsheet but who find current commercially available OLAP packages inadequate as they have very convoluted and limited rules languages for expressing business logic, especially, financial or corporate models.

7 15

Sydney scoreboard?
by Sarah Burke Jan. 8, 2000

Jan. 8, 2000

I think I remember reading on the web site a while ago that there was a scoreboard in a Sydney stadium programmed in Python. Does anyone know anything more about this? Thanks

9 12